A Salinity Module for SWAT to Simulate Salt Ion Fate and Transport 2 at the Watershed Scale 3 4 Ryan T

Hydrol. Earth Syst. Sci. Discuss., https://doi.org/10.5194/hess-2018-614 Manuscript under review for journal Hydrol. Earth Syst. Sci. Discussion started: 22 January 2019 c Author(s) 2019. CC BY 4.0 License.

1 A Salinity Module for SWAT to Simulate Salt Ion Fate and Transport 2 at the Watershed Scale 3 4 Ryan T. Bailey1*, Saman Tavakoli-Kivi1, Xiaolu Wei1 5 1 Department of Civil and Environmental Engineering, Colorado State University, 1372 Campus Delivery, Fort Collins, CO, 6 80523-1372, United States. 7 8 *Correspondence to: Ryan Bailey ([email protected]) 9 10 Abstract. Salinity is one of the most common water quality threats in river basins and irrigated regions worldwide. However, no 11 available numerical models simulate all major processes affecting salt ion fate and transport at the watershed scale. This study

12 presents a new salinity module for the SWAT model that simulates the fate and transport of 8 major salt ions (SO4, Ca, Mg, Na,

13 K, Cl, CO3, HCO3) in a watershed system. The module accounts for salt transport in surface runoff, soil percolation, lateral flow, 14 groundwater, and streams, and equilibrium chemistry reactions in soil layers and the aquifer. The module consists of several new 15 subroutines that are imbedded within the SWAT modelling code and one input file containing soil salinity and aquifer salinity 16 data for the watershed. The model is applied to a 732 km2 salinity-impaired irrigated region within the Arkansas River Valley in 17 southeastern Colorado, and tested against root zone soil salinity, groundwater salt ion concentration, groundwater salt loadings to 18 the river network, and in-stream salt ion concentration. The model can be a useful tool in simulating baseline salinity transport 19 and investigating salinity best management practices in watersheds of varying spatial scales worldwide. 20 21 1 Introduction 22 Salinity is one of the most common water quality threats in river basins and irrigated regions worldwide. Sustainability of 23 crop production in irrigated areas in semi-arid and arid areas is threatened by over-irrigation, poor quality of irrigation water 24 (high salinity), inadequate drainage, shallow saline groundwater, and salinization of soil and underlying groundwater, all of 25 which can lead to decreasing crop yield. Of the estimated 260 million ha of irrigated land worldwide, approximately 20-30 26 million ha (7-12%) is salinized (Tanji and Kielen, 2002), with a loss of 0.25 to 0.5 million ha each year globally. Approximately 27 8.8 million ha in western Australia alone may be lost to production by the year 2050 (NLWRA, 2001), and 25% of the Indus 28 River basin is affected by high salinity. Within the western United States, 27-28% of irrigated land has experienced sharp 29 declines in crop productivity due to high salinity (Umali, 1993; Tanji and Kielen, 2002), thereby rendering irrigated-induced 30 salinity as the principal water quality problem in the semi-arid regions of the western United States. 31 Salinization of soil and groundwater systems is caused by both natural processes and human-made activities. Salt naturally

32 can be dissolved from parent rock and soil material, with salt minerals (e.g. gypsum CaSO4, halite NaCl) dissolving to mobile 2+ - + - 33 ions such as Ca , SO4 , Na , and Cl . In addition, salt ions can accumulate in the shallow soil zone due to waterlogging, which is 34 a result of over-irrigating and irrigating in areas with inadequate drainage. Salts moving up into the soil zone can become evapo- 35 concentrated due to the removal of pure water by crop roots. Soil water salinization leads to a decrease in osmotic potential, i.e. 36 the potential for water to move from soil to the crop root cells via osmosis, leading to a decrease in crop production. 37 Numerical models have been used extensively to assess saline conditions, simulate salt movement across landscapes and 38 within soil profiles, predict salt build-up and movement in the root zone, and investigate the impact of best management 39 practices (Oosterbaan, 2005; Schoups et al., 2005; Burkhalter and Gates, 2006; Singh and Panda, 2012). Available models that 40 either have inherent salinity modules or can be applied to salinity transport problems include UNSATCHEM (Šimůnek and 1

41 Suarez, 1994), HYDRUS linked with UNSATCHEM (Šimůnek et al., 2012); DRAINMOD, LEACHC (Wagenet and Hutson, 42 1987), SAHYSMOD (Oosterbaan, 2005; Singh and Panda, 2012), CATSALT, and MT3DMS (Burkhalter and Gates, 2006). 43 Whereas several of these models include major ion chemistry for salt ions (e.g. precipitation-dissolution, cation exchange, 44 complexation) (UNSATCHEM, HYDRUS), their application typically is limited to small field-scale or soil-profile domains (e.g. 45 Kaledhonkar and Keshari, 2006; Schoups et al., 2006; Kaledhonkar et al., 2012; Rasouli et al., 2013). Conversely, models such 46 as SAHYSMOD and MT3DMS have been applied to regional-scale problems, but lack the reaction chemistry and treat salinity 47 as a conservative solute. SAHYSMOD uses seasonal water and salt balance components for large-scale systems on a seasonal 48 time step (Singh and Panda, 2012). MT3DMS is a finite-difference contaminant transport groundwater model that uses 49 MODFLOW output for groundwater flow rates, but does not include salt ion solution chemistry (Burkhalter and Gates, 2006). 50 Schoups et al. (2005) used a hydro-salinity model that couples MODHMS with UNSATCHEM to simulate subsurface salt 51 transport and storage in a 1,400 km2 region of the San Joaquin Valley, California. The model, however, does not consider 52 salinity transport in surface runoff or salt transport in streams, limiting results to soil salinity and groundwater. Currently, there is 53 no model that simulates salt transport in all major hydrologic pathways (surface runoff, soil percolation and leaching, 54 groundwater flow, streamflow) at the watershed-scale that also considers important solution reaction chemistry. Such a model is 55 important for assessing watershed-scale and basin-scale salt movement and investigating the impact of large-scale salinity 56 remediation schemes. 57 The objective of this paper is to present a salinity transport modeling code that can be used to simulate the fate and transport

58 of the major ions (SO4, Ca, Mg, Na, K, Cl, CO3, HCO3) in a watershed hydrologic system. The salinity module is implemented 59 within the SWAT modeling code, and thereby salt transport pathways include surface runoff, percolation, soil later flow, 60 groundwater flow and streamflow. The soil water and groundwater concentration of each salt ion is also affected by equilibrium 61 chemistry reactions: precipitation-dissolution, complexation, and cation exchange. The use of the model is demonstrated through 62 application to a 732 km2 region of the Lower Arkansas River Valley (LARV) in southeastern Colorado, an irrigated alluvial 63 valley in which soil and groundwater salinization has occurred over the past few decades. The model is tested against salt ion and 64 total dissolved solids (TDS) concentration in surface water (Arkansas River and its tributaries), groundwater (from a network of 65 monitoring wells), and soil water (from a large dataset of soil salinity measurements). The salinity module for SWAT can be 66 applied to any watershed to simulate baseline conditions and to test the effect of best management practices on watershed 67 salinity. 68 69 2 Development of the SWAT Salinity Module 70 This section provides a brief overview of the SWAT model, followed by a description of the SWAT salinity module. Sect. 3 71 demonstrates the use of the salinity module to a regional-scale irrigated stream-aquifer system in the Lowe Arkansas River 72 Valley, Colorado. 73 2.1 The SWAT Model 74 The SWAT (Soil and Water Assessment Tool, Arnold et al., 1998) hydrologic model simulates water flow, nutrient 75 mass transport and sediment mass transport at the watershed scale. It is a continuous, daily time-step, basin-scale, distributed- 76 parameter watershed model that simulates water flow and nutrient (nitrogen, phosphorus) transport in surface runoff, soil 77 percolation, soil later flow, groundwater flow and discharge to streams, and streamflow. The watershed is divided into subbasins, 78 which are then further divided into multiple unique combinations (Hydrologic Response Units HRUs) of land use, soil type and 79 topographic slope for which detailed water and nutrient mass balance calculations are performed. Routing algorithms route water 80 and nutrient mass through the stream network to the watershed outlet. SWAT has been applied to hundreds of watersheds and

81 river basins worldwide to assess water supply and nutrient contamination under baseline conditions (Abbaspour et al., 2015) and 82 scenarios of land use change (Zhao et al., 2016; Zuo et al., 2016; Napoli et al., 2017), best management practices (Arabi et al., 83 2006; Maringanti et al., 2009; Ullrich and Volk, 2009; Dechmi and Skhiri, 2013), and climate change (Jyrkama and Sykes, 2007; 84 Ficklin et al., 2009; Tweed et al., 2009; Haddeland et al., 2010; Brown et al., 2015). However, it has not yet been applied to 85 salinity issues. 86 2.2 Salinity Module for SWAT

87 The new SWAT salinity module simulates the fate and transport of 8 major salt ions (SO4, Ca, Mg, Na, K, Cl, CO3, HCO3) 88 via surface runoff, soil later flow, soil percolation and leaching, groundwater flow, and streamflow, subject to chemical reactions 89 such as precipitation-dissolution, complexation, and cation exchange within soil layers and the alluvial aquifer. The module also 90 simulates the loading of salt mass to the soil profile via saline irrigation water from both surface water (subbasin channel) and 91 groundwater (aquifer) sources. A watershed cross-section schematic describing these processes is shown in Figure 1. 92 The salinity module is implemented directly into the SWAT FORTRAN code, with new subroutines developed for salt 93 chemistry (salt_chem), salt irrigation loading (salt_irrig), salinity percolation and leaching (salt_lch), and salt groundwater 94 transport and loading to streams (salt_gw). Other standard SWAT subroutines are modified to incorporate salt ion transport and 95 effects, such as SWAT’s crop growth modules, lagging solutes in surface runoff and groundwater flow (surfstor, substor), and 96 routing solutes through the stream network (watqual). These subroutines are shown in Figure 2 within the general SWAT 97 modeling code data flow. For each day loop, the mass balance calculations for each HRU are performed. Salt subroutines are 98 shown for chemical equilibrium, irrigation loading, salt leaching, soil salinity stress, salt groundwater transport and loading, and 99 lagging in surface runoff and groundwater flow. At the end of the HRU calculations, the water, sediment, nutrients, and salt mass 100 is routed through the stream network, with in-stream concentration of each salt ion simulated for each SWAT subbasin. Details 101 for each salt ion process are now presented. For the equations presented, S refers to salt mass, and the subscript i refers to the 8 102 major ions. For the transport equations, calculations are similar to SWAT’s transport equations for nitrate. Salinity module input 103 data and output data also will be discussed later in this section. 104 2.2.1 Salt in Surface Runoff (“salt_lch” and “surfstor” subroutines) 105 The mass of each salt ion can be transferred from an HRU to the subbasin channel via surface runoff. The salt ion mass

' 106 generated in surface runoff Sisurf, (kg/ha) for the current day is calculated as:

107 SCQ'  (1) i, surf Sii S surf

108 where  is the salinity percolation coefficient, C is the concentration of the ith salt ion in the mobile water for the top 10 mm Si Si

109 of soil (kg salt /mm water), and Qsurf is the surface water generated from the HRU on a given day (mm water). As only a portion 110 of the surface runoff and lateral flow reaches the subbasin channel on the day it is generated, SWAT uses a storage feature to 111 surface runoff. The salt ion mass reaching the subbasin channel on the current day via surface runoff is calculated as:

surlag 112 SSS' 1exp (2) i,,, surf i surf i surfstor  tconc

th 113 where Si,surf is the mass of the i salt ion that reaches the subbasin channel on the current day (kg/ha), Si, surfstor is the salt ion

114 surface runoff stored or lagged from the previous day (kg/ha), surlag is the surface runoff lag coefficient, and tconc is the time of 115 concentration for the HRU (hrs). 116 2.2.2 Lateral Flow (“salt_lch” and “substor” subroutines)

' 117 The salt ion mass generated in lateral flow Silatly,,(kg/ha) from a soil layer for the current day is calculated as:

118 SCQ'  (3) i,, lat ly Si lat , ly

119 where Qlat,ly is the water discharge from the layer by lateral flow (mm water). Similar to surface runoff, only a portion of the 120 lateral flow will reach the subbasin channel on the day it is generated, and thus the salt ion mass reaching the channel on the

121 current day Silatly,,(kg/ha) via lateral flow is calculated as:

1 122 SSS' 1exp (4) ilatly,, ilatly ,, ilatstor ,  TTlat

123 where Silatstor, is the salt ion mass stored or lagged from the previous day (kg/ha) and TTlag is the lateral flow travel time (days). 124 2.2.3 Soil Percolation (“salt_lch” subroutine) 125 The salinity module tracks the mass of each salt ion (kg/ha) in each soil layer. The salt ion mass moved to the underlying

126 soil layer by percolation Sipercly,,(kg/ha) is calculated as: 127 SCQ (5) ipercly,, Si percly ,

128 where Qlat,ly is the amount of water percolating to the underlying soil layer on a given day (mm water). After percolation has been 129 simulated, the concentration of each salt ion (mg/L) in each soil layer is calculated using the area (m2) of the HRU and the 130 volume of water in the soil layer (m3). The leached salt ion mass is added to the shallow aquifer using the following: 131 SgwSgwS1    (6) i,,,,1 rech delay i perc delay i rech t  where is the salt ion mass loaded to the water table via recharge (kg/ha), is the salt ion mass percolated from the 132 Si, rech Siperc,

133 bottom layer of the soil profile, Si,,1 rech t is the leached salt ion mas from the previous day, and gwdelay is the groundwater delay 134 time, i.e. the time required for water leaving the bottom of the root zone to reach the water table (days). 135 2.2.4 Groundwater Flow (“salt_gw” subroutine) 136 The salinity module tracks the mass of each salt ion (kg/ha) in the aquifer. The salt ion mass generated in groundwater flow 137 ' (kg/ha) from the aquifer for the current day is calculated as: Sigw, 138 SCQ'  (7) igw, Sigw, gw

139 whereC is the salt ion concentration in the aquifer (kg salt /mm water), andQ is the groundwater flow generated for the HRU Sigw, gw 140 for the current day (mm water). The concentration of each salt ion in each HRU aquifer is calculated on each day by dividing the 141 total mass of the salt ion (g) by the total volume of groundwater (m3). 142 2.2.5 Streamflow (“watqual” subroutine) 143 Water is routed through the watershed channel network using the variable storage routing method, a variation of the 144 kinematic wave model (Neitsch et al., 2011). The mass of each salt ion is routed through the channel network with water, with no 145 chemical reactions changing in-stream salt ion concentration. Similar to any constituent in SWAT, salt ion loadings (kg/day) can 146 be specified for any subbasin reach of the watershed. 147 2.2.6 Salt Loading in Irrigation water (“salt_irrig” subroutine) 148 Salt ion mass is added to the soil profile via irrigation water, with water derived from either the aquifer (groundwater 149 pumping) or from surface water diversions. Including constituent mass in irrigation water is a new feature for SWAT, as the 150 original code does not account for nutrient (N, P) mass in irrigation water. If the irrigation water source is a subbasin reach 151 (surface water irrigation), the concentration of each salt ion is multiplied by the volume of applied irrigation water (depth of 152 water * HRU area) to determine the mass of each salt ion (kg/ha) to add to the first soil layer. If the irrigation water source is the 4

153 shallow aquifer, the concentration of each salt ion in the HRU aquifer is used to estimate salt loading to the first soil layer. The 154 salt ion mass is then removed from the HRU aquifer. 155 2.2.7 Salt Solution Chemistry 156 The salinity chemistry implemented into SWAT is based on the Salinity Equilibrium Chemistry (SEC) module developed 157 for soil-aquifer systems (Tavakoli-Kivi, 2018). The equations for salinity solution chemistry presented here are performed for 158 each HRU soil layer and for each HRU. The solution chemistry in this module is similar to that implemented in other water 159 chemistry models [UNSATCHEM: Šimůnek et al. (2012), PHREEQC: Parkhurst and Appelo (2013), MINTEQA2: Paz-Garcia 160 et al. (2013)]. Thus, only basic details are presented here. 161 The SEC module includes 8 aqueous components, 10 complexed species, five solid (salt mineral) species, and four exchange

162 species (Table 1). The 8 aqueous components (SO4, Ca, Mg, Na, K, Cl, CO3, HCO3) are included due to their presence in the

163 majority of soil-aquifer systems. The five salt minerals (CaSO4, CaCO3, MgCO3, NaCl, MgSO4) also are included due to their 164 presence in many soil-aquifer systems, although the module can be amended to include any mineral species. The module 165 simulates the dissolved concentration (mg/L) of the 8 ions in soil water and groundwater and the solid mass concentration of the 166 five salt mineral species in the soil and the aquifer sediment according to precipitation-dissolution, complexation, and cation 167 exchange reactions. 168 For these calculations, the duration of the model time step (daily time step for SWAT) is assumed long enough for all 169 constituent reactions to achieve equilibrium. The concentration of species at equilibrium is calculated using a stoichiometric 170 algorithm approach, in which mass balance and mass action equations are solved simultaneously. This method is used in other 171 water chemical equilibrium packages such as PHREEQC (Parkhurst and Appelo, 2013) and MINTEQA2 (Paz-Garcia et al., 172 2013). 173 Law of Mass Action 174 At equilibrium, the concentration of all reactants and products are related using the equilibrium constant K: (C)cd (D) 175 K  (8) (A)ab (B) 176 where A and B are reactants, C and D are reactants, a, b, c, and d are constants, and the parentheses denote solute activities. The th 177 activity of the i solute, iA, is computed by multiplying the activity coefficient γi by the molal concentration, where γi depends on 178 the ionic strength I of the solution: 1 179 Imz . 2 (9) 2  ii th 180 where zi is the charge number of the i ion and mi is the molality (mol/kg H20). γi is then given as:  2 Azai I log i I  0.1 181  1 Baai I (10)  I log Az2  0.3 I 0.1  I  0.5  ii  1 I -1 10 -1 o 182 where Aa and Ba are temperature dependent constants (Aa = 0.5085 m and Ba = 0.3285×10 m at 25 C) and ai is a measure of 183 effective diameter of a hydrated ion i. The first equation in (10) is the Debye-Huckle equation for dilute solutions, and the second 184 equation is the Davis equation. 185 Mass Balance Equations 186 The mass of each element in the system, either in ion or complexed form, is tracked by a set of mass balance equations.

187 Equations for SO4, Cl, Ca, and Na are:

2- 0 0 - - 188 SO =[SO ]+[CaSO ]+[MgSO ]+[NaSO ]+[KSO ] (11a) 44T 4 4 4 4

- 189 ClT =[Cl ] (11b)

2+ 0 0 + 190 CaT =[Ca ]+[CaSO43 ]+[CaCO ]+[CaHCO 3 ] (11c)

+- - 0 191 NaT =[Na ]+[NaSO43 ]+[NaCO ]+[NaHCO 3 ] (11d)

192 where T denotes total concentration and brackets indicate species’ molality. Similar equations are written for Mg, K, CO3, and

193 HCO3. 194 Precipitation-Dissolution Reactions

195 Salt minerals (ABs) can dissolve or precipitate according to the stoichiometric reaction

+- 196 ABsaqaq A + B (12)

+  197 The salt mineral will dissolved if the solution is under-saturated in in regards to A aq and B aq , and will precipitate if the

198 solution is super-saturated. Salt minerals in the SEC module include CaSO4, CaCO3, MgCO3, MgSO4, and NaCl, due to their 199 common occurrence in aquifers. For example:

2+ 2- 200 CaSO44 Ca + SO (13) 201 with a solubility product constant: (Ca2+ )(SO 2- ) 202 4 (14) Ksp  CaSO4 (CaSO4 ) 203 Within the SEC module, minerals are added to the system one at a time, with the solubility limits of each mineral used to 204 determine the direction of each reaction (precipitation or dissolution). 205 Complexation Reactions 206 Based on the law of mass action, equilibrium equations are written for all complexed species. For example, the equation for

0 207 CaSO4 is:

(Ca22 )(SO ) 208 K  4 (15) CaSO4 0 CaSO4

209 where K is the equilibrium constant and is equal to 0.004866. Equations and equilibrium constants for the remaining 9 CaSO4 210 complexed species are shown in Supporting Material. 211 Cation Exchange Reactions 212 Cation exchange is calculated to determine the sorbed and released ions from sediment surfaces to the solution. The order of 213 replaceability is Na > K > Mg > Ca, determined by Coulomb’s Law. The cation reaction as an equivalent reactions represented 214 by Gapon equation:

nm 215 XnNXmX1/mM 1/ 1/ nN 1/ (16)

216 where X1/mM is exchangeable cation M on the surface (meq/100), X1/nN is exchangeable cation N on the surface (meq/100g), M and 217 N are metal cations, and m+ and n+ are the charges of cations M and N respectively. Using the cation exchange capacity of the 218 soil and a coefficient of Gapon selectivity coefficient for each reaction, concentration of each exchangeable species is 219 determined. 220 221 6

222 2.2.8 Salinity Module Input/Output 223 Required data for running the SWAT salinity module include: precipitation-dissolution solubility products for the five salt

224 minerals (CaSO4, CaCO3, MgCO3, NaCl, MgSO4), initial concentration of salt ions in soil water and groundwater, and initial salt 225 mineral solid concentration (% of bulk soil) in soil and aquifer sediment. Initial concentrations are required for each HRU. 226 However, as will be shown in Sect. 3, using uniform (i.e. all HRU values are the same) concentration values yields the same 227 result as using spatially-variable initial concentrations, if a warm-up period of several years is used in the SWAT simulation. 228 All input data are provided in a single input file, “salt_input”. To turn on the salinity module, a single line has been added at 229 the end of the file.cio file, with flag being read (0 or 1) to exclude/include the salinity module. If the flag is set to 1, the SWAT 230 code will open and read the contents of the salt_input file. 231 232 3 Application of SWAT Salinity Module to an Irrigated Stream-Aquifer System 233 3.1 Study Region: Lower Arkansas River Valley, Colorado 234 The salinity module is tested for a 732 km2 irrigated stream-aquifer system along the Arkansas River in southeastern 235 Colorado (Figure 3A). The region consists the Arkansas River and tributaries (e.g. Timpas Creek, Crooked Arroyo, see Figure 236 3A) running through and over a thin (~10-15 km in width) and shallow (~10-20 m) sandy alluvial aquifer. The climate is semi- 237 arid, requiring irrigation to supplement rainfall for crop growth. Irrigation water is derived either from the Arkansas River via a 238 system of irrigation canals or from the aquifer via a network of ~500 pumping wells (Figure 3A). Cultivation and associated 239 irrigation occurs March through November. 240 Salinization of soil, groundwater, and surface water in the region has steadily worsened since the 1970s due to increased 241 irrigation diversions from the Arkansas River, high water tables due to excessive water applications to fields, and the existence

242 of salt minerals, particularly gypsum (CaSO4) (Konikow and Person, 1985; Goff et al., 1998; Gates et al., 2002; Gates et al., 243 2016). Soil salinity levels under about 70% of the area exceed threshold tolerance for crops, with the regional average of crop 244 yield reduction from salinity and waterlogging estimated to range from 11 to 19% (Gates et al., 2002; Morway and Gates, 2012). 245 From sampling groundwater from a network of 82 observation wells (see Figure 3B) (sampling from June 2006 to May 246 2010), average salinity concentration of shallow groundwater is approximately 2,700 to 3,000 mg/L, and annual salt loading to 247 the Arkansas River from groundwater return flows is about 500 kg per irrigated ha, per km of the river. In the 1990s, 68% of 248 producers stated that high salinity levels are a significant concern (Fraser et al., 1999). For the region modeled in this study,

249 average TDS concentration ( CTDS ) in groundwater is 3,334 mg/L (443 samples), with a minimum of 459 mg/L and a maximum

250 of 44,600 mg/L. The presence of gypsum is revealed in the high concentration of SO4 ( C ), with average, minimum, and SO4 251 maximum concentrations of 1,878 mg/L, 147 mg/L, and 29,457 mg/L, respectively. Average soil salinity, using electrical 252 conductivity (EC), is 4.11 dS/m (54,700 measurements), with minimum and maximum of 0.9 dS/m and 56.5 dS/m, respectively. 253 Based on 6 surface water sampling sites (4 in the Arkansas River, 2 in tributaries; Figure 3B), average C and C is 1145 TDS SO4 254 mg/L and 560 mg/L, respectively. More details of observed groundwater, soil water, and surface water concentrations are 255 provided in Sect. 3.3.2 when model results are presented. 256 3.2 SWAT Model 257 A previously calibrated and tested SWAT model for the study region is used to simulate salt fate and transport using the 258 developed salinity module. The SWAT model is detailed in Wei et al. (2018). The region was divided into 72 subbasins (see 259 Figure 3B). A method was developed to apply SWAT to highly-managed irrigated watersheds, and included: designating each 260 cultivated field as an individual HRU (see Figure 3B for the map of fields); crop rotations to simulate the effects of changing

261 crop types for each field during the 11-year simulation; seepage to the aquifer from the earthen irrigation canals; and SWAT’s 262 auto-irrigation algorithms to trigger irrigation events based on plant water demand for both surface water irrigation and 263 groundwater irrigation. The method resulted in 5,270 HRUs. Implementing canal seepage required a slight change to the SWAT 264 modeling code to add pre-processed, estimated canal seepage to HRU aquifer. Canal seepage rates were obtained from field 265 measurements (Susfalk eta l., 2008; Martin et al., 2014). The model was run for the 1999-2009 time period, with simulated 266 streamflow compared to observed hydrographs at 5 stream gages (Rocky Ford, La Junta, Las Animas, Timpas Creek, Crooked 267 Arroyo; see Figure 3B) for model testing (Wei et al., 2018). 268 3.3 SWAT Model with Salinity Module 269 3.3.1 Model Construction and Simulation 270 The SWAT model is run from April 1 1999 to December 13 2009, with observed data for testing available from June 2006 271 to December 2009. The 1999-2005 period thus serves as a warm-up simulation period. The calibration period is 2006-2007, and 272 the testing period from 2008-2009. Required inputs include initial soil water and groundwater ion concentrations, initial soil and 273 aquifer sediment salt mineral fractions and, due to the study region being a part of the larger Lower Arkansas River Valley, ion 274 mass loading in the Arkansas River at the upstream end of the modeled region (Catlin Dam; see Figure 3B). 275 Salt ion mass loading (kg/day) in the Arkansas River at Catlin Dam were estimated using daily measured values of EC 276 (dS/m) and streamflow (m3/s) and periodic measurements of salt ion concentration (mg/L). Linear relationships were established 277 between EC and the concentration of each salt ion, with this relationship then used to estimate salt ion concentration for each day 278 of the simulation period. The daily in-stream mass of each salt ion was then calculated by multiplying daily salt ion 279 concentration by streamflow, and added to the point-source SWAT input file for the appropriate subbasin. Figure 4A shows the 280 daily loading (kg/day) for each salt ion using this method. The make-up of total mass loading by salt ion is shown in Figure 4B,

281 with SO4 accounting for 47% of total in-stream salt mass. The linear relationship between EC and selected salt ions (SO4, Cl, Na) 282 and TDS is shown in the charts along the bottom of Figure 4. For TDS the R2 value of the relationship is approximately 0.93. 283 Initial salt ion concentrations in soil water and groundwater were based on averages of observed groundwater 284 concentrations. For the baseline simulation, the same values were assigned to each HRU. These are 1875 mg/L, 330 mg/L, 175 285 mg/L, 440 mg/L, 10 mg/L, 150 mg/L, 5 mg/L, and 350 mg/L for C , C , C , C , C , C , C , and C , SO4 Ca Mg Na K Cl CO3 HCO3 286 respectively. The effect of using spatially-varying initial concentrations is explored in additional scenarios. Salt mineral fractions

287 for CaSO4 and CaCO3 in the HRU soil layers are based on a soil survey of the region from the Natural Resources Conservation

288 Service (NRCS). The fraction of soil that is CaSO4 and CaCO3 was set to 0.1 and 0.01. For the aquifer sediment, fractions are 289 based on the spatial patterns determined in Tavakoli-Kivi (2018) for a salinity groundwater transport study of the same region. 290 Solubility products for precipitation-dissolution of salt minerals were obtained from literature and from Tavakoli-Kivi (2018) -9 -6 -5 291 and are 3.07 x 10 , 4.8 x 10 , 4.9 x 10 , 0.0072, and 37.3 for CaCO3, MgCO3, CaSO4, MgSO4, and NaCl, respectively, for both 292 soil and aquifer sediments. 293 Only minimal manual calibration was applied to the model, to yield correct magnitudes of salt ion concentration in soil

294 water, groundwater, and stream water. Targeted parameters were the solubility product of CaSO4 precipitation-dissolution, and

295 the soil fraction of CaSO4. The solubility produce was increased from 0.000049 to 0.0003, and the soil fraction of CaSO4 was 296 decreased from 0.01 to 0.009. Model results are tested against in-stream concentration of salt ions, soil water EC (dS/m), 297 groundwater concentration of salt ions, and groundwater salt ion mass loading to the Arkansas River. Observed soil EC values 298 were obtained using a saturated paste extract, and hence comparison with model results will not be as rigorous as for 299 groundwater and surface water data.

300 Several variations of the model were run to test the effect of 1) initial salt ion concentrations and 2) specified loading of salt 301 ion mass at the upstream end of the Arkansas River. For 1), the variations include uniform initial concentrations (baseline 302 model), random spatially-variable concentrations, and initial concentrations equal to 0. For 2), the variation included one 303 simulation with no loading. 304 3.3.2 Model Results 305 Model results consist of in-stream salt ion and TDS concentration, hydrologic pathway (groundwater discharge, surface 306 runoff, percolation) salt loadings, groundwater salt ion concentration, soil water EC, watershed-wide salt balance, and 307 groundwater salt loading to the Arkansas River. 308 3.3.2.1 In-Stream Salt Ion Concentration 309 Simulated and observed in-stream salt ion concentrations (mg/L) are shown in Figure 5 for the Rocky Ford site (Figure 5A)

310 and the Crooked Arroyo site (Figure 5B). Results are shown for SO4, Ca, Cl, and HCO3, with the calculated Nash-Sutcliffe 311 model efficiency coefficient (NSE) shown on each plot. Results for TDS at all 5 gaging stations are shown in Figure 6. As can be 312 seen by the trends in concentration and also the NSE values, the SWAT model performs well in replicating in-stream salt ion

313 concentrations, particularly for SO4 (NSE = 0.60), Ca (NSE = 0.54), HCO3 (NSE = 0.73), and TDS (NSE = 0.69) in the Arkansas 314 River at the Rocky Ford gaging site. The model does not perform as well in downstream sites, with NSE at La Junta and at Las 315 Animas equal to 0.34 and 0.25, respectively, although the trends are correct and the magnitudes are correct except for at the 316 downstream-most site (Las Animas), where the model under-predicts total salt concentration. This is also shown by a 1:1 317 comparison of all salt ion data for the Rocky Ford (Figure 7A) and Las Animas (Figure 7C) sites, which yield R2 values of 0.87 318 and 0.74, respectively. Las Animas also has an R2 value of 0.74. However, as the SWAT model often is used to estimate monthly 319 in-stream loads rather than daily in-stream concentration, these results are promising regarding the use of SWAT to estimate in- 320 stream salinity loadings. 321 In regards to the NSE, the model performs rather poorly in the two tributaries (Timpas Creek, Crooked Arroyo), with NSE 322 equal to -0.32 and 0.41, respectively, for TDS (Figure 6B, 6C). However, the overall trends and magnitude compare well to 323 observed data. This is shown in the 1:1 plot of all salt ion data for Timpas Creek in Figure 7B, resulting in an R2 value of 0.79. 324 The relationship for Crooked Arroyo yields an R2 value of 0.80. This is particularly promising given that there is no specified 325 upstream loading for the tributaries, and hence all salt mass within the stream system is due to surface runoff, lateral flow, and 326 groundwater discharge. Hence, comparing simulated and observed in-stream salinity concentration in these two systems is a 327 strong test for the model. 328 Figure 8 shows the salt loading via the hydrologic pathways of groundwater discharge (Figure 8A), surface runoff (8B), and 329 percolation from the soil profile to groundwater (8C). For Timpas Creek, 96% of salt in the creek water is from groundwater 330 discharge, 3% from surface runoff, and 1% from lateral flow. For Crooked Arroyo, the portions are 91%, 6%, and 3%, and for 331 the Arkansas River they are 96%, 3%, and 1%, highlighting the strong influence of groundwater on surface water salt load. This 332 is shown further by examining the domain-wide salt balance, presented in Sect. 3.3.2.3. The mass loading of total salt from the 333 aquifer to the Arkansas River for each day of the 2006-2009 time period is shown in Figure 9. Mass balance plot values are the 334 mean of a a stochastic river mass balance calculation of surface water salinity loadings along the length of the Arkansas River 335 within the model domain, using a method similar to Mueller-Price and Gates (2008), with values indicating the mass of salt not 336 accounted for by surface water loadings. These unaccounted for loadings include groundwater, and thus provide an upper limit of 337 in-stream salt loading from groundwater discharge. 338 339

340 3.3.2.2 Groundwater and Soil Water Salinity 341 Groundwater salt results are shown by spatial maps and by comparison of frequency distributions. For all simulated results, 342 only concentration values from days on which field samples were taken are included in the analysis. Time-averaged TDS (mg/L),

343 SO4 (mg/L), and Na (mg/L) in groundwater is shown for each HRU in Figure 10. Also shown is soil water EC (dS/m) for each

344 HRU soil profile, and the percent of the soil profile (Figure 10E) and aquifer (Figure 10F) that is CaSO4 (solid mineral) at the 345 end of the simulation period. These maps are shown to provide an indication of the degree of spatial variation simulated by the

346 salinity module. Variation in each system response is large, with TDS ranging from 0 to ~11,700 mg/L, SO4 from 0 to ~6700 347 mg/L, and Na from 0 to ~1,270 mg/L. In comparison, if data from an outlier monitoring well are excluded (monitoring well with

348 salinity values more than double of any other monitoring well), the maximum observed values for TDS, SO4, and Na are 13,000 349 mg/L, 6,500 mg/L, and 2,600 mg/L. 350 Results for all salt ions are summarized in Table 2. Average concentration of field samples (based on field samples from 82

351 monitoring wells shown in Figure 3B) and HRU-simulated groundwater salinity compares well, particularly for SO4 (1,878 mg/L 352 to 2,058 mg/L) and for TDS (3,334 mg/L to 3,276 mg/L). In addition to a comparison of maximum and average values,

353 comparison at various magnitude levels is performed using relative frequency plots, shown in Figure 11. Results for SO4 (Figure

354 11A), HCO3 (11B), and TDS (11C) are shown. Similar to the results shown in Table 2, the comparison for SO4 and TDS is good,

355 but the model generally under-predicts HCO3 for most HRUs. A relative frequency plot of observed and simulated EC (dS/m) in 356 the soil profile also is shown (Figure 11D). The average of observed values and simulated values are 4.1 dS/m and 4.8 dS/m, 357 although the majority of observed values are between 2 dS/m and 4 dS/m whereas no such grouping occurs for the simulated 358 values. However, the observed data values are obtained from saturated paste extracts, which therefore lowers the salinity 359 concentration due to the addition of water to bring the soil to saturation. Hence, the “observed” (modified by the saturated paste 360 method) concentrations should be lower than what actual occurs in the field, which may explain the disagreement shown in 361 Figure 11D. 362 3.3.2.3 Salt Balance 363 The domain-wide salt balance is presented in Figure 12A. All salt balance components are included, with all values scaled 364 according to the small salt flux (lateral flow = 1 unit). For the soil profile, salt is added via groundwater irrigation (12 units), 365 surface water irrigation (33), dissolution of salt minerals (110), and upflux from the aquifer saturated zone (39), and removed via 366 percolation (103), surface runoff (4), and lateral flow (1). A similar salt balance can be performed for each salt ion in the system. 367 Salt removed from the aquifer and added to the soil profile via upflux is approximately 30% of percolation, which compares well 368 to a comparison of water upflux and recharge magnitudes computed by Morway et al. (2013) in a groundwater modeling study of 369 the region using MODFLOW. 370 Of the salt entering the river, 96.7% is from groundwater (151 units out of 156), and the remaining from surface runoff and 371 lateral flow. Time series of daily loading (kg/ha) for these three components is shown in Figure 12B, and loadings for 372 percolation, surface water irrigation, and groundwater irrigation are shown in Figure 12C, showing the seasonal trends in 373 applying irrigation water. These results also indicate that much of the salt leaching from the soil profile is due to dissolution of 374 salt minerals. Results also indicate the importance of including salt mass in applied irrigation water, as it accounts for 375 approximately half of salt leaching to the aquifer. Finally, results show the importance of including precipitation-dissolution in 376 the module, as this process is a large component of the salt balance. Without including this process, the module would severely 377 under-predict salt ion concentrations throughout the watershed, demonstrating the need to include each salt ion individually as 378 opposed to modeling salinity as a conservative solute in the system. 379

380 381 3.3.2.4 Scenarios and Model Guidelines 382 The effect of initial salt ion concentrations and upstream salt ion mas loading is summarized by the time series charts in

383 Figure 13. For the Rocky Ford and Las Animas gaging sites, a time series of simulated SO4 (mg/L) and TDS (mg/L) is compared 384 for the following scenarios: uniform initial salt ion concentration (“Original”: this refers to the baseline simulation); HRU- 385 variable initial concentration (“Variable IC”); initial concentrations equal to 0 (“Zero IC”); and not accounting for upstream salt 386 ion mass loading at Catlin Dam (“No US Loading”). 387 There are only small differences between using uniform or HRU-variable initial concentrations for soil water and 388 groundwater. Any differences are readily resolved during the warm-up period. Hence, to facilitate model use we recommend that 389 uniform initial concentrations be used. 390 Using initial concentrations equal to 0 mg/L has a significant effect, particularly for downstream sites such as Las Animas 391 (Figure 13C, D). For this watershed, salt loading to the streams is principally from groundwater, and if soil water and 392 groundwater are not provided with initial salt ion concentrations, the groundwater salt ion loading to subbasin streams is small 393 compared to the baseline simulation. As downstream flow and in-stream salt loading is effected by groundwater loading, these 394 areas (e.g. Las Animas site) experience the effect more acutely than upstream sites such as Rocky Ford (Figure 13A,B). 395 However, by the end of the simulation (2009), difference between “Zero IC” and “Original” is small. This is shown by the “Diff” 396 time series for each plot. Therefore, if groundwater discharge is a large component of total water yield for the watershed, “Zero 397 IC” should not be used, or a long warm-up simulation period needs to be used. 398 Not including upstream salt ion loading at Catlin Dam has a stronger effect on the Rocky Ford site (Figure 13A,B) than at 399 the outlet (Las Animas) (Figure 13C,D). This is due to Las Animas being much farther downstream, and hence there is much 400 more groundwater salt ion loading to the streams that can make up for the salt not included at the upstream end of the Arkansas 401 River at Catlin Dam. Overall, any point sources of in-stream salt should be added, unless only downstream areas are targeted for 402 baseline simulations and best management practice investigation. The effect of neglecting point sources of in-stream salt 403 decreases as the groundwater loading component of total salt yield increases. 404 The importance of including equilibrium chemistry into the salt transport module is demonstrated by the results shown in 405 Figure 14. The simulated in-stream TDS (mg/L) is shown at the Rocky Ford site (Figure 14A), the Timpas Creek site (B), and 406 the Las Animas site (C), for both the original simulation (red line) and a simulation “No SEC” that does not include the SEC 407 module (black line). The “No SEC” simulation therefore represents a system wherein salt is transported through the stream- 408 aquifer system as a conservative species. Clearly, in-stream concentrations are much too low for the simulation without the SEC 409 module. This is due to the neglect of salt mineral dissolution, which in the actual system transfers salt mass from the soil and 410 aquifer material to soil water and groundwater are thereby increases the loading of salt to the stream network. For this system, 411 and likely most watersheds, equilibrium chemistry must be included to establish the correct magnitude of salt loading and 412 concentrations. 413 3.3.3 Model Use and Limitations 414 The salinity module of SWAT differs from other salinity models in that it accounts for salt loading for each major 415 hydrologic pathway in a watershed setting (stream, groundwater, lateral flow, surface runoff, tile drain flow), for each major salt 416 ion, subject to chemical equilibrium reactions (precipitation-dissolution, complexation, cation exchange). As such, it can be used 417 to estimate baseline salt loading within a watershed, and also explore the impact of land management and water management 418 scenarios to mitigate soil salinity, groundwater salinity, and surface water salinity. The model, however, does not simulate 419 physically-based, spatially-distributed groundwater flow and solute transport with an accurate depiction of water table elevation

420 and groundwater head gradient, and thus the trends in groundwater salt loading to streams may not be accurate (see Figure 9). To 421 overcome this issue, the new salinity module could be incorporated into SWAT-MODFLOW (Bailey et al., 2016), which links 422 SWAT and MODFLOW to simulate land surface and subsurface flow processes, and SWAT-MODFLOW-RT3D (Wei et al., 423 2018), which includes reactive transport of solutes into SWAT-MODFLOW. 424 425 4 Conclusions 426 This study presents a new watershed-scale salt ion fate and transport model, by developing a salinity module for the SWAT 427 model. The module accounts for salt loading for each major hydrologic pathway in a watershed setting (stream, groundwater,

428 lateral flow, surface runoff, tile drain flow), for each major salt ion (SO4, Ca, Mg, Na, K, Cl, CO3, HCO3). The module also 429 accounts for principal equilibrium chemistry reactions (precipitation-dissolution, complexation, cation exchange). For

430 precipitation-dissolution, five salt minerals (CaSO4, CaCO3, MgCO3, NaCl, MgSO4) have been included. The model was applied 431 and tested in a 732 km2 irrigated stream-aquifer watershed in southeastern Colorado, along the alluvial corridor of the Arkansas 432 River. Model results are tested against in-stream salt ion concentration, groundwater salt ion concentration, soil salinity, and 433 groundwater salt loading to the Arkansas River. 434 The model can be used to assess baseline salinity conditions in a watershed and to explore land and water management 435 strategies aimed at decreasing salinization in river basins. Such strategies may include on-farm management, lining irrigation 436 canals to reduce saline canal seepage, dry-drainage practices, and reducing volumes of applied irrigation water. Due to the 437 simulation of soil water salt ion concentrations and SWAT’s simulation of crop growth, the salinity module can also be used to 438 investigate the effect of these strategies on crop yield. Although this study applied the model to an irrigated area, the model can 439 be applied to non-irrigated areas as well. 440 441 Code Availability 442 The code consists of the original SWAT files, with 6 additional files for the salinity module. All files are *.f FORTRAN files. 443 The code is available at the following URL: https://github.com/rtbailey8/SWAT_Salinity/tree/v1.0.0 (DOI: 444 10.5281/zenodo.2541224). An example model input file (salt_input) and example output files are also provided. 445 446 Author Contribution 447 Ryan Bailey wrote the salinity module for SWAT and tested the module for the study region. Saman Tavakoli-Kivi prepared the 448 solution chemistry algorithms for the salinity module. Xiaolu Wei prepared and tested the original SWAT model for the study 449 region, and facilitated use of the new salinity module for the constructed SWAT model. 450 451 Competing Interests 452 The authors declare that they have no conflict of interest. 453 454 References 455 Abbaspour, K.C., Rouholahnejad, E., Vaghefi, S., Srinivasan, R., Yang, H., and Klove, B.: A continental-scale hydrology and 456 water quality model for Europe: Calibration and uncertainty of a high-resolution large-scale SWAT model. Journal of 457 Hydrology 524, 733-752, https://doi.org/10.1016/j.jhydrol.2015.03.027, 2015.

458 Arabi, M., Govindaraju, R.S., Hantush, M.M., and Engel, B.A.: Role of watershed subdivision on modeling the effectiveness of 459 best management practices with SWAT. Journal of the American Water Resources Association 42(2), 513-528, 460 https://doi.org/10.1111/j.1752-1688.2006.tb03854.x, 2006. 461 Bailey, R.T., Wible, T.C., Arabi, M., Records, R.M., and Ditty, J.: Assessing regional-scale spatio-temporal patterns of 462 groundwater-surface water interactions using a coupled SWAT-MODFLOW model. Hydrological Processes 30, 4420-4433, 463 https://doi.org/10.1002/hyp.10933, 2016. 464 Brown, S.C., Versace, V.L., Lester, R.E. and Walter, M.T.: Assessing the impact of drought and forestry on streamflows in 465 south-eastern Australia using a physically based hydrological model, Environmental Earth Sciences, 74(7), 6047-6063, 466 available: http://dx.doi.org/10.1007/s12665-015-4628-8, 2015. 467 Burkhalter, J. P. and Gates, T. K.: Evaluating regional solutions to salinization and waterlogging in an irrigated river valley. 468 Journal of Irrigation and Drainage Engineering, 132(1): 21 – 30, https://doi.org/10.1061/(ASCE)0733- 469 9437(2006)132:1(21), 2006. 470 Dechmi, F. and Skhiri, A.: Evaluation of best management practices under intensive irrigation using SWAT model. Agricultural 471 Water Management 123, 55-64, https://doi.org/10.1016/j.agwat.2013.03.016, 2013. 472 Ficklin, D.L., Luo, Y., Luedeling, E. and Zhang, M.: Climate change sensitivity assessment of a highly agricultural watershed 473 using SWAT, Journal of Hydrology, 374(1-2), 16-29, available: http://dx.doi.org/10.1016/j.jhydrol.2009.05.016, 2009. 474 Frasier, W.M., Waskom, R.M., Hoag, D.L., and Bauder, T.A.: Irrigation Management in Colorado: Survey Data and Findings. 475 Technical Report TR99-5 Agricultural Experiment Station, Colorado State University. Fort Collins, CO, 1999. 476 Gates, T.K., Burkhalter, J.P., Labadie, J.W., Valliant, J.C. and Broner, I.: Monitoring and modeling flow and salt transport in a 477 salinity-threatened irrigated valley, Journal of Irrigation and Drainage Engineering, Vol. 128, No. 2, 88-99, 478 https://doi.org/10.1061/(ASCE)0733-9437(2002)128:2(87), 2002. 479 Gates, T.K, Steed, G.H., Niemann, J.D. and Labadie, J.W.: Data for Improved Water Management in Colorado’s Arkansas River 480 Basin, Hydrological and Water Quality Studies, Colorado State University, 2016. 481 Goff, K., M.E. Lewis, M.A. Person, and Konikow, L.F.: Simulated effects of irrigation on salinity in the Arkansas River Valley 482 in Colorado. Ground Water 36:76-86, https://doi.org/10.1111/j.1745-6584.1998.tb01067.x, 1998. 483 Haddeland, I., Heinke, J., Voss, F., Eisner, S., Chen, C., Hagemann, S. and Ludwig, F.: Effects of climate model radiation, 484 humidity and wind estimates on hydrological simulations, Hydrology and Earth System Sciences, 16(2), 305-318, available: 485 http://dx.doi.org/10.5194/hess-16-305-2012, 2012. 486 Jyrkama, M.I. and Sykes, J.F.: The impact of climate change on spatially varying groundwater recharge in the grand river 487 watershed (Ontario), Journal of Hydrology, 338(3-4), 237-250, available: http://dx.doi.org/10.1016/j.jhydrol.2007.02.036, 488 2007. 489 Kaledhonkar, M. J. and Keshari, A. K.: Regional salinity modeling for conjunctive water use planning in Kheri command. J. 490 Irrig. Drain. Eng., 132(4), 389–398, https://doi.org/10.1061/(ASCE)0733-9437(2006)132:4(389), 2006. 491 Kaledhonkar, M.J., Sharma, D.R., Tyagi, N.K., Kumar, A., and Van Der Zee, S.E.A.T.M.: Modeling for conjunctive use 492 irrigation planning in sodic groundwater areas. Agricultural Water Management 107, 14-22, 493 https://doi.org/10.1016/j.agwat.2011.12.023, 2012. 494 Konikow, L.F. and Person, M.: Assessment of long-term salinity changes in an irrigated stream aquifer system. Water Resources 495 Research 21:1611-1624, https://doi.org/10.1029/WR021i011p01611, 1985.

496 Maringanti, C., Chaubey, I. and Popp, J.: Development of a multiobjective optimization tool for the selection and placement of 497 best management practices for nonpoint source pollution control. Water Resources Research, 498 doi.org/10.1029/2008WR007094, 2009. 499 Martin, C.A. and Gates, T.K.: Uncertainty of canal seepage losses estimated using flowing water balance with acoustic Doppler 500 devices. Journal of Hydrology 517: 746-761, https://doi.org/10.1016/j.jhydrol.2014.05.074, 2014. 501 Morway, E.D. and Gates, T.K.: Regional assessment of soil water salinity across an intensively irrigated river valley, Journal of 502 Irrigation and Drainage Engineering, Vol. 138, No. 5, 393-405, https://doi.org/10.1061/(ASCE)IR.1943-4774.0000411, 503 2012. 504 Morway, E.D., Gates, T.K. and Niswonger, R.G.: Appraising options to reduce shallow groundwater tables and enhance flow 505 conditions over regional scales in an irrigated alluvial aquifer system. Journal of Hydrology 495, 216-237, 506 https://doi.org/10.1016/j.jhydrol.2013.04.047, 2013. 507 Mueller Price, J. and Gates, T.K.: Assessing uncertainty in mass balance calculation of river nonpoint source loads. Journal of 508 Environmental Engineering, 134(4), 247-258, https://doi.org/10.1061/(ASCE)0733-9372(2008)134:4(247), 2007. 509 Napoli, M., Massetti, L, and Orlandini, S.: Hydrological response to land use and climate changes in a rural hilly basin in Italy. 510 CATENA 157, 1-11, https://doi.org/10.1016/j.catena.2017.05.002, 2017. 511 Neitsch, S.L., J.G. Arnold, J.R. Kiniry, and Williams, J.R.: Soil and Water Assessment Tool Theoretical Documentation, Version 512 2009. Temple, Tex.: Texas Water Resources Institute Technical Report No. 406, 2011. 513 National Land and Water Resources Audit, 2001, Australian Government. 514 Oosterbaan, R. J.: SAHYSMOD (version 1.7a), Description of principles, user manual and case studies, International Institute 515 for Land Reclamation and Improvement, Wageningen, Netherlands, 140, 2005. 516 Parkhust, D.L. and Appelo, C.A.J.: Description of Input and Examples for PHREEQC Version 3- A Computer Program for 517 Speciation, Batch-Reaction, One-Dimensional Transport, and Inverse Geocehmical Calculations, Chapter 43 of Section A, 518 Groundwater, Book 6, Modeling Techniques, 2013. 519 Paz-García J., Johannesson, B., Ottosen, L., Ribeiro, A., Rodríguez-Maroto, J.: Computing multi-species chemical equilibrium 520 with an algorithm based on the reaction extents, Computers & Chemical Engineering, Vol: 58 pp: 135-143, 521 https://doi.org/10.1016/j.compchemeng.2013.06.013, 2013. 522 Rasouli, F., Pouya, A.K., and Šimůnek, J.: Modeling the effects of saline water use in wheat-cultivated lands using the 523 UNSATCHEM model. Irrigation Science 31(5), 1009-1024, https://doi.org/10.1007/s00271-012-0383-8, 2013. 524 Schoups, G., Hopmans, J.W., Young, C.A., Vrugt, J.A., Wallender, W.W., Tanji, K.K., and Panday, S.: Sustainability of 525 irrigated agriculture in the San Joaquin Valley, California. Proceedings of the National Academy of Sciences of the United 526 States of America, 102(43), 15352-15356, https://doi.org/10.1073/pnas.0507723102, 2005. 527 Schoups, G., Hopmans, J.W. and Tanji, K.K.: Evaluation of model complexity and space–time resolution on the prediction of 528 long‐term soil salinity dynamics, western San Joaquin Valley, California. Hydrological Processes, 20(13), 2647-2668, 529 https://doi.org/10.1002/hyp.6082, 2006. 530 Šimůnek, J., and Suarez, D.L.: Two-dimensional transport model for variably saturated porous media with major ion chemistry, 531 Water Resources Research, 30(4), 1115-1133, https://doi.org/10.1029/93WR03347, 1994. 532 Šimůnek, J., M. Šejna, and van Genuchten, M. Th.: The UNSATCHEM Module for HYDRUS (2D/3D) Simulating Two- 533 Dimensional Movement of and Reactions Between Major Ions in Soils, Version 1.0, PC Progress, Prague, Czech Republic, 534 54 pp, 2012.

535 Singh, A., and Panda, S.N.: Integrated salt and water balance modeling for the management of waterlogging and salinization; I: 536 Validation of SAHYSMOD, Journal of Irrigation and Drainage Engineering, Vol. 138, 955-963, 537 https://doi.org/10.1061/(ASCE)IR.1943-4774.0000511, 2012. 538 Susfalk R, Sada D, Martin C, Young MH, Gates T, Rosamond C, Mihevc T, Arrowood T, Shanafield M, Epstein B, and 539 Fitzgerald B.: Evaluation of linear anionic polyacrylamide (LA-PAM) application to water delivery canals for seepage 540 reduction. Desert Research Institute. DHS Publication No. 41245, 2008. 541 Tavakoli-Kivi, S.: Simulating the fate and transport of salinity species in a semi-arid agricultural groundwater system: model 542 development and application. PhD Dissertation, Dept. of Civil and Environmental Engineering, Colorado State University, 543 Fort Collins, CO, 2018. 544 Tanji, K. and Kielen, N.: Agricultural drainage water management in arid and semi-arid areas, FAO Irrigation and Drainage 545 Paper 61, Rome, 2002. 546 Tweed, S., Leblanc, M. and Cartwright, I.: Groundwater-surface water interaction and the impact of a multi-year drought on 547 lakes conditions in South-East Australia, Journal of Hydrology, 379(1-2), 41-53, 548 http://dx.doi.org/10.1016/j.jhydrol.2009.09.043, 2009. 549 Ullrich, A. and Volk, M.: Application of the Soil and Water Assessment Tool (SWAT) to predict the impact of alternative 550 management practices on water quality and quantity. Agricultural Water Management 96(8), 1207-1217, 551 https://doi.org/10.1016/j.agwat.2009.03.010, 2009. 552 Umali, D.L.: Irrigation– induced salinity: A growing problem for development and the environment: Technical Paper 215. World 553 Bank, Washington, DC, 1993. 554 Wagenet, R.J., and Hutson, J.L.: LEACHM-Leaching estimation and chemistry model. Center Environ. Res., Cornell Univ., 555 Ithaca, NY, 1987. 556 Wei, X., Bailey, R.T. and Tasdighi, A.: Using the SWAT model in intensively managed irrigated watersheds: model 557 modification and application. Journal of Hydrologic Engineering 23(10): 04018044, 558 https://doi.org/10.1061/(ASCE)HE.1943-5584.0001696, 2018. 559 Zhao, A., Zhu, X., Liu, X., Pan, Y., and Zuo, D.: Impacts of land use change and climate variability on green and blue water 560 resources in the Weihe River Basin of northwest China. CATENA 137, 318-327, 561 https://doi.org/10.1016/j.catena.2015.09.018, 2016. 562 Zuo, D., Xu, Z., Yao, W., Jin, S., Xiao, P., and Ran, D.: Assessing the effects of changes in land use and climate on runoff and 563 sediment yields from a watershed in the Loess Plateau of China. Science of the Total Environment 544, 238-250, 564 https://doi.org/10.1016/j.scitotenv.2015.11.060, 2016. 565 566 567 568 569 570 571 572 573 574

575 576 577 578 579 580

581 582 Figure 1. Schematic showing a cross-section of an irrigated stream-aquifer system and the major transport pathways of salt, 583 which consists of the eight major ions of SO4, Ca, Mg, Na, K, Cl, CO3, HCO3. The concentration of each ion is also governed by 584 equilibrium chemistry reactions such as precipitation-dissolution, complexation, and cation exchange within the soil profile and 585 within the aquifer. 586

587 588 Figure 2. Data flow within the SWAT-Salt modeling code. Boxes and text in black and blue indicate original SWAT loops and 589 subroutines. Text in red indicates either new or modified subroutines for the Salinity module. The required input data for the 590 salinity module is shown in the upper shaded box, whereas the generated output files are shown in the lower shaded box. 591 592

593 594 Figure 3. Map of study region within the Lower Arkansas River Valley of Colorado, showing (A) Arkansas River and 595 tributaries, irrigation canals, and pumping wells, and (B) cultivated fields, monitoring wells where groundwater is sampled for 596 salt ions, sampling sites where surface water is sampled for salt ions, and SWAT subbasins. 597 598 599 600

601 602 Figure 4. Data summarizing the specified loading of salt (kg/day) at the Catlin Dam gage site, using observed EC (dS/m) and 603 stream discharge (m3/day) data: (A) daily loading of salt ion, (B) percentage of total salt loading attributed to each salt ion, 604 (bottom charts) example regression plots used to relate EC to salt ion concentration. 605 606 607 18

608 609 610 Figure 5. Time series of simulated and observed concentration (mg/L) of selected salt ions for the (A) Rocky Ford sampling site 611 along the Arkansas River (see Fig. 3) and the (B) Crooked Arroyo sampling site. The Nash-Sutcliffe model efficiency coefficient 612 (NSE) is shown for each plot. 613

614 615 Figure 6. Simulated and observed total dissolved solids (TDS) (mg/L) in the five stream sampling sites along the Arkansas River 616 (A, D, E), and two tributaries (B, C). See Fig. 3 for locations. TDS is the summation of the concentration of the 8 salt ions. The 617 Nash-Sutcliffe model efficiency coefficient (NSE) is shown for each plot. 618 619

620 621 Figure 7. Log-log plots of observed vs. simulated salt ion concentration for the (A) Rocky Ford, (B) Timpas Creek, and (C) Las 622 Animas surface water sampling sites. (D) shows the comparison of TDS for the five sites. 623 624 625 626

627 628 Figure 8. Average daily loading (kg/ha) of salt by subbasin to (A) stream network via groundwater discharge, (B) stream 629 network via surface runoff, (C) groundwater via soil percolation. 630

631 632 Figure 9. Simulated daily mass loading of TDS (kg) to the Arkansas River via groundwater discharge for the SWAT model with 633 uniform initial salt concentrations. Results from a salt mass balance calculation on the Arkansas River also are plotted, showing 634 the unaccounted for TDS loadings (groundwater, surface runoff, small inflows) in the Arkansas River. 635 636 637 638 639

640 641 Figure 10. HRU average concentration over the 2006-2009 simulation period for (A) groundwater TDS (mg/L), (B) groundwater 642 SO4 (mg/L), (C) groundwater Na (mg/L), and (D) soil water electrical conductivity EC (dS/m). (E) and (F) show percentage of 643 soil bulk volume and aquifer bulk volume, respectively, that is CaSO4, near the end of the simulation in May 2010. 644 645 646

647

648 649 Figure 11. Relative frequency plots of simulated and observed values of (A) SO4 groundwater concentration, (B) HCO3 650 groundwater concentration, (C) TDS groundwater concentration, and (D) EC soil water concentration. Simulated values are 651 taken from each HRU of the SWAT simulation, on days for which observed values are available. 652 653 654 655

656 657 Figure 12. Magnitude of salt balance components in the watershed model for TDS, showing (A) relative salt flux between soil 658 storage compartments in the watershed for each salt transport pathway; (B) daily loading (kg/ha) of salt in groundwater, surface 659 runoff, and lateral flow to streams; and (C) daily loading (kg/ha) of salt in percolation water (from bottom of soil profile to the 660 aquifer), irrigation derived from irrigation canals, and irrigated derived from groundwater pumping. 661 662

663 664 665 Figure 13. Simulated in-stream SO4 and TDS concentration (mg/L) at the Rocky Ford Site and the Las Animas Site along the 666 Arkansas River for four scenarios: uniform initial conditions (IC) of salt soil water and groundwater concentrations, 667 corresponding to the original simulation; variable IC; IC = 0; and no upstream loading of salt at the Catlin Dam site. Also show 668 is the difference between the IC = 0 scenario and the original scenario. 669 670 671

672 673 Figure 14. Simulated in-stream TDS concentration (mg/L) at the (A) Rocky Ford Site, (B) Timpas Creek Site, and (C) Las 674 Animas Site for the original simulation (red line) and a simulation without including equilibrium chemistry (SEC module) (black 675 line).

676 Table 1. Groups and Species included in the Salinity Equilibrium Chemistry (SEC) module for SWAT.

Group Species 2+ 2+ + + -2 2- - - Aqueous Species Ca , Mg , Na , K ,SO43 , CO , HCO 3 ,Cl Solid Species CaSO4,CaCO3,MgCO3,NaCl, MgSO4 000 + 0 CaSO443 , MgSO , CaCO , CaHCO 3 , MgCO 3 , Complexed Species MgHCO+-- , NaSO , KSO , NaHCO 0 , NaCO 0 344 33 Exchanged Species Ca, Mg, Na, K 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 25

712 Table 2. Summary statistics for observed (monitoring well) and simulated (SWAT) salinity concentrations in groundwater. 713 Maximum (mg/L) Average (mg/L) Species Observed Simulated Observed Simulated Na 2606 1269 402 187 Ca 767 2234 353 653 Mg 1019 497 191 78 K 85 277 4 9

SO4 6510 6738 1878 2058

CO3 42 8 2 0

HCO3 2362 1828 410 225 Cl 1803 480 95 65 TDS 13007 11667 3334 3276 714 715